My Learnings After Writing 100s of Prompts for LLMs
here are my learning after writing 100s of prompts for llms :
1. learn the vocab before blindly prompting, sometimes researching the exact words for your usecase/niche is useful in getting a quality response for eg : instead of asking for a analysis of your business in detail, search how to analyse businesses, and put in something like, provide me a detailed “SWOT” Analysis of my business
2. treat it like your friend, tell it how you’re equal partners in whatever you do, and if you benefit, i benefit from your response. 3. it’s biased, make it forget that. tell it to be practical and act in the real world, tell it to act like a monk while thinking, but still being rational enough
4. feed it what it knows, use transformations, for example, if you need something in a very weird format you just invented, break it down — ask for the formats it knows (JSON, Typespec, Pydantic, YAML) — tell it your personal-weird-format (not judging lol) — transform the json from original to your format
5. before starting a prompt, almost always give it a role, find a role in which a human can do your work, find someone famous and give it that role, if not name, a description can be enough research and hint, but know what you’re looking for : quirky and fun writing : ask for
type writings rather than plain quirky
— — here are some tips from my own experiences, i use these, for my own work and also for the specific-llm, i use the prompt instructions given by the developers like OpenAI, Anthropic etc, it helps a lot — -
Thanks a lot
Hitesh Bandhu — https://x.com/_hiteshbandhu
